EP2761329A2

EP2761329A2 - Method for validating a training image for the multipoint geostatistical modeling of the subsoil

Info

Publication number: EP2761329A2
Application number: EP12745502.0A
Authority: EP
Inventors: Cyril RONOT; Philippe RUELLAND; Pierre Biver
Original assignee: Total SE
Current assignee: TotalEnergies SE
Priority date: 2011-09-30
Filing date: 2012-07-24
Publication date: 2014-08-06
Anticipated expiration: 2032-07-24
Also published as: US20140225890A1; FR2980898A1; FR2980897B1; EP2761340A2; WO2013045776A3; EP2761329B1; FR2980898B1; US10209401B2; WO2013045776A2; FR2980897A1; EP2761340B1; WO2013045777A3; WO2013045777A2; US9279905B2; US20140244227A1

Abstract

According to the invention, a training image denotes, for each cell of a first three-dimensional grid, a respective type of geological facies assigned to said cell. The representativeness of the training image is verified with respect to data obtained from measurements and/or samplings carried out in a plurality of wells drilled into the area of the subsoil in question. For each cell of a second three-dimensional grid, the morphology of which is matched to estimated geological horizons, the well data comprise a respective type of geological facies estimated for said cell. Two-dimensional patterns, each of which is defined along a respective geological horizon and includes, in a cell containing the intersection of a well with said geological horizon, the type of geological facies estimated for each cell, are defined. The instances of the identified two-dimensional patterns are counted in the training image so as to verify the ability of the training image to account for correlations between wells.

Description

DRIVE IMAGE VALIDATION METHOD FOR THE MODELING G EOSTATI STI THAT MULTIPOINT OF THE BASEMENT

The present invention relates to multipoint geostatistical techniques used to build geological models to represent areas of the subsoil.

[0002] Geological models are very useful tools for the study of the subsoil, in particular for the exploration or exploitation of hydrocarbon fields. We try to build a geological model that best represents the structure and behavior of a reservoir from measurements made either on the surface (including seismic measurements) or in wells (logs), or from observations made on carrots taken from the wells. Of course, even when large amounts of measurements and samples are taken, the construction of a three-dimensional model very accurately reflecting the structure and behavior of the studied sub-soil area is not a realistic goal. This is why statistical methods are used to construct these models.

[0003] Traditional geostatistical methods rely on the use of variograms which are two-point spatial variability measurements. These traditional methods are very limited, especially as to their ability to represent the sometimes complex topology of the study area. Multipoint methods have been proposed since the 1990s to overcome these limitations.

[0004] Multipoint Point Statistics (MPS) techniques make it possible to model geological heterogeneities in such a way that natural systems can be reproduced by better accounting for the uncertainties of the model.

[Q005] The input data of MPS simulation techniques include training images, the choice of which plays an important role in the reliability of the model obtained. A training image is defined on a three-dimensional grid (3D). It is chosen to represent patterns of geological heterogeneities that can be expected to occur in the reservoir (see S. Strebelle, "Mathematical Geology, Vol 34, No. 1, 2002, pp. 1-21).

In the early MPS techniques, we used a stationary training image forming a conceptual representation of a sedimentary system, obtained from a sketch prepared by a geologist or from a simulation based on objects. More recent algorithms have enabled MPS simulations using non-stationary training images, combining them with auxiliary training images (see TL Chugunova & LY Hu, "Multiple-Point Simulations Constrained"). Continuous Auxiliary Data, "Mathematical Geosciences, Vol 40, No. 2, 2008, pp. 133-146).

The training images can in particular correspond to a simplified representation of a contemporary analogue, obtained by an image processing technique applied to photographs taken by satellite in a region of the globe which, in the opinion of Geological experts, has characteristics similar to those of the area where sedimentary deposits were formerly formed in the studied area of the subsoil. For example, different satellite images are processed and stacked to form a three-dimensional representation that will serve as a training image. Very often, the study of a particular geological area involves the drilling of one or more wells in this area to make measurements (logs) and / or samples. It is then possible to check whether the 3D training image seems consistent with the available well data, in order to validate the training image or indicate that it should be modified. Indeed, a training image that is not consistent with the well data is likely to provide a poor representation of the heterogeneities during the simulation.

[0009] Methods applicable to the validation of vertical heterogeneities on training images have been proposed by JB Boisvert, J. Pyrcz and CV Deutsch, "Multiple-Point Statistics for Training Image Selection", Natural Resources Research, Vol. 16, No. 4, December 2007, pp. 313-320. Multipoint statistics such as the Multipoint Density Function (MPDF, Multi- Point Density Function ") are calculated according to a dimension based on the well data and on the training image.If there is a good match between the two MPDF functions, we deduce that the training image is consistent with the well data An object of the present invention is to enrich the methods of generating and validating training images, in particular it is desirable to be able to validate a training image with respect to well data without being limited to statistical estimates in parallel with a well.

According to the invention, there is provided a method for preparing a training image for modeling a subsoil area by a multipoint geostatistical technique. This process comprises:

constructing a training image on a first three-dimensional cell grid, the training image giving, for each cell of the first three-dimensional grid, a respective type of geological facies assigned to this cell; and

- verify the representativity of the training image with respect to well data obtained from observations made in several wells that have been drilled in the basement area, the well data referring to a second grid of three-dimensional cells whose morphology is adapted to geological horizons estimated in the subsoil zone, the well data including types of geological facies respectively estimated from the observations for cells of the second three-dimensional grid located along the well.

The representativeness check of the drive image with respect to the well data comprises an inter-well verification which comprises:

identifying two-dimensional patterns in the well data, each two-dimensional pattern being defined along a respective geological horizon and including, in a cell containing the intersection of a well with said geological horizon in the second three-dimensional grid, the type estimated geological facies for this cell; and

counting, in the training image, the occurrences of the two-dimensional patterns identified. The representativeness check of the drive image with respect to the well data is not limited to a vertical verification. A lateral verification is performed to exploit the information provided by the correlations from one well to another, by means of an analysis based on two-dimensional patterns that can be identified for groups of M neighboring wells, where M is a chosen number greater than 1 . This also increases the probability that the model that will be generated by the MPS technique adequately represents the studied area of the subsoil.

The inter-well verification may comprise a validation of the training image when the number of counted occurrences of the identified two-dimensional patterns exceeds a threshold, which threshold may be expressed proportionally to the number of possible positions of the two-dimensional pattern in the training picture.

In one embodiment, each two-dimensional pattern, defined for wells crossing a geological horizon, is identified by the position of the cell of the second three-dimensional grid containing the barycenter of these wells along the geological horizon considered. Counting, in the training image, the occurrences of a two-dimensional pattern identified by the position of a centroid cell may then comprise:

scanning the training image, each step of the scanning placing the two-dimensional pattern on the first three-dimensional grid by aligning the cell containing the center of gravity of the wells on a reference cell of the first three-dimensional grid and the cells of the two-dimensional pattern respectively containing the intersections between the wells and the geological horizon on target cells of the first three-dimensional grid; and

incrementing a counter each time a scan step matches the facies types estimated for the cells of the two-dimensional pattern with the facies types respectively assigned to the target cells in the training image.

The representativeness check of the training image with respect to the well data may be supplemented by an intra-well verification comprising: - estimating a multipoint statistical measure, such as a multipoint density function (MPDF) or a distribution of runs ("distribution of runs"), along each well among the well data;

estimating the same multipoint statistical measure along vertical columns of the first three-dimensional cell grid; and

- compare the two estimated multipoint statistical measures.

Another aspect of the present invention relates to a data processing system adapted to the preparation of a training image for modeling a subsoil area by a multipoint geostatistical technique. The system includes computer resources configured to implement a method as defined above.

Yet another aspect of the present invention relates to a computer program for a data processing system, the program comprising instructions for implementing a method as defined above when executed on a computing unit of the data processing system. A computer readable recording medium on which such a program is recorded is also provided.

Other features and advantages of the present invention will appear in the following description of a nonlimiting exemplary embodiment, with reference to the accompanying drawings, in which:

- Figures 1 and 2 are schematic illustrations of a regular grid and a tank grid, respectively;

FIG. 3 is a flowchart of an exemplary embodiment of a method according to the invention;

FIGS. 4A-4E show examples of pretreated satellite images for assembling a three-dimensional training image;

FIGS. 5 to 7 are diagrams illustrating the assembly of the 3D training image;

FIGS. 8A-B are diagrams illustrating a first example of multipoint statistical measurement; FIG. 9 is a diagram illustrating a second example of multipoint statistical measurement;

FIG. 10 illustrates a way of checking the representativity of the drive image with respect to well data;

FIG. 11 is another illustration of the verification of representativity of the training image with respect to the well data;

FIG. 12 is a flowchart, an exemplary procedure applicable in an inter-well check of a training image;

FIGS. 13 to 15 are diagrams illustrating a technique making it possible to take into account well connectivity information in the initialization of an MPS simulation; and

Fig. 16 is a flowchart showing one way to implement the technique illustrated in Figs. 13-15.

Multipoint statistical techniques (MPS) have developed over the last twenty years. The facies of the geological zone considered are modeled pixel by pixel on a grid of three-dimensional cells whose morphology is adapted to the geological horizons in the zone to be modeled. This 3D grid is called tank grid. The MPS simulation assigns to each pixel of the reservoir grid a facies value, for example an integer, corresponding to a codification of some types of facies that are expected to be found in the zone to be modeled. For example, in a fluvial environment, facies types may be channel, alluvial plain, sandy deposit, crevasse splay, etc.

The implementation of an MPS technique requires having a 3D drive image whose statistical content is representative of the area to be modeled. This training image is typically defined on a first grid of three-dimensional cells of regular shape, and contains statistical information on sedimentary facies on the scale of geological bodies ("geobody").

MPS simulation may consist of scanning the training image using a rectangular window, to identify the different facies configurations encountered and organize them into a list or a search tree. Then, from the initially filled cells of the reservoir grid (for example from samples taken in wells), the algorithm traverses the reservoir grid and retrieves the compatible configurations in the search tree and then draws facies values. from these configurations according to their probabilistic distribution function. He thus constructs, step by step, a reservoir model whose spatial correlations of facies are statistically similar to those of the training image.

Of course, the quality of the modeling strongly depends on the ability of the training image to adequately represent the statistics of the modeled area.

An even grid of cells as illustrated in Figure 1 is a rectangular parallelepiped-shaped three-dimensional matrix containing a number of rectangular parallelepiped shaped cells of the same size. Each cell is referenced by its position in a reference I, J, K, where I and J are horizontal indexes and K represents a layer index in a set of layers superimposed in the vertical direction.

A reservoir grid (Figure 2) is a 3D grid whose topology reproduces that of a reservoir located in an area of the basement. The shape of the reservoir grid is defined by horizons that have been recorded on two or three dimensional seismic images. The techniques of pointing seismic horizons in 2D or 3D seismic images are well known. The horizons correspond to subsoil surfaces where a strong seismic impedance contrast is observed. Horizons are related to changes in lithology and often correspond to stratification plans. The tectonic activity of the Earth deformed these initially flat surfaces, and formed faults and / or folds. The reservoir grid, using these surfaces as a frame, takes into account the geological structures. The cells of the reservoir grid are referenced by their position in a reference frame Ir, Jr, Kr, where Kr represents a layer index and Ir, Jr is the position index of the cells inside each layer. Figure 2 shows a relatively simple example of deformation in a reservoir grid. In practice, the layers often have more complex deformations, such as those of the layer shown in Figure 11 (a). In general, the number of layers in the grid of the drive image is less than the number of layers in the reservoir grid.

An embodiment of the drive image preparation method for the MPS simulation of the subsoil zone is presented with reference to FIG. 3. In a first step (steps 10-1 1), 2D images are obtained, such as those shown in Figures 4A-E. Then a 3D expansion step 12 is performed to assemble a drive image, as shown in FIGS. 5-7.

In the example of FIG. 3, obtaining the 2D images begins with the digitization of 10 satellite images that have been taken in a suitable region of the globe. This region is chosen by geological experts for its supposed similarity, in terms of sedimentary behavior, with that where the geological area studied was formed.

The semi-automatic digitization of satellite images by image processing techniques makes it possible to quickly extract the parts of a photograph that are significant in terms of the sedimentary deposit body. These image processing techniques are based on algorithms such as Canny's Edge Detection (Canny, A Computational Approach to Edge Detection, IEEE Transactions on Pattern Analysis and Machine Intelligence, PAMI-8, No. 6, November 1986, pp. 679-698) or segmentation by a watershed transform (J. Roerdink & A. Meijster, "The Watershed Transform: Definitions, Algorithms and Parallelization Strategies", Fundamenta Informaticae, Vol 41, 2002, pp. 187-228). A 2D grid is created and filled with discrete facies values interpreted from the satellite images using an image processing algorithm as mentioned above.

In step 12, the digitized 2D images are stacked in the vertical direction (Figure 5). The result is (Figure 6) a 3D image to which another processing, able to use a pseudo-process algorithm, is applied to extrapolate the 3D geological body shape. A 3D drive image is thus obtained, such as that illustrated in FIG. 7, comprising for example of the order of 3 to 5 layers in the vertical direction K and a few hundred cells in each horizontal direction I, J.

This training image is tested in step 13 to verify stationarity, that is to say if it has a spatial uniformity in terms of orientation or facies proportion. If the image is not stationary, auxiliary training images representing spatial trends are generated from the training image. Auxiliary training images may include drifts regarding facies proportions, azimuths, geological body sizes. They are built semiautomatically in step 14, for example according to the technique described in the aforementioned article by T. L. Chugunova & L.Y. Hu, and then allow to perform the MPS simulation in step 18 despite the non-stationarity of the drive image.

In step 15 of FIG. 3, the 3D training image generated in step 13 is evaluated with respect to data available at one or more wells that have drilled in the reservoir zone. and in which measurements and samples have been taken. A number of statistical consistency criteria between the training image and the well data are evaluated. If one or more of these criteria are poorly verified, the training image is rejected in step 16, and another training image will be generated, for example by modifying the previous training image or starting over again. a method 10-14 similar to that described above. This other training image will be tested again until a satisfactory training image is obtained against the evaluated criteria. Another possibility at the end of step 15 is to present the user with the evaluation results of the criteria taken into account, to let the user decide on the results if the training image can be accepted. or not for MPS simulation.

Step 15 may include an intra-well check relating to the multipoint statistics in parallel with the wells. This can notably be done on the basis of the distribution of runs, as illustrated by FIGS. 8A-B, or the multipoint density function (MPDF), as illustrated by FIG. 9 (cf. the aforementioned article by JB Boisvert, J. Pyrcz and CV Deutsch).

Series distribution is a statistical measure of the vertical extent of facies along a well. For each facies, we count the number consecutive blocks along the wells that have this facies. By block is meant here the number of layers of the reservoir grid which have the facies considered at their intersection with the well. For example, along the well schematized in Figure 8A, there is 1 series of 3 blocks, 2 sets of 2 blocks and 4 sets of 1 block. The numbers recorded are noted on a diagram such as that of Figure 8B, where the size of the series is plotted on the abscissa. The same reading is made along the vertical columns of the training image defined on the regular grid of cells. The two diagrams obtained are compared and the distance between them gives an indication of the capacity of the training image to account for the vertical extent of the facies along the wells. If the distance is too large, the training image may be rejected.

The Multipoint Density Function (MPDF) is another multipoint statistical measure along the wells that can be estimated at step 15 on the one hand from the well data on the reservoir grid and on the other hand from the training image along the vertical columns of the regular grid of cells. MPDF is a count of the number of occurrences of different 1D facies patterns along the wells. Figure 9 gives a simplified representation in the case of patterns on three layers among two types of facies. A comparison of the number of occurrences of the different patterns along the wells and in the training image thus makes it possible to evaluate the statistical relevance of the training image with a view to informing the user or rejecting the image. drive if necessary.

Step 15 may further include an inter-well check whose purpose is to verify that the drive image has, along its layers I, J, two-dimensional facies patterns that correspond to patterns. two-dimensional observations observed in the wells along the geological horizons Ir, Jr of the reservoir grid.

To perform this inter-well check, we first identify two-dimensional patterns in the well data. Each two-dimensional pattern is defined along any geological horizon. In a cell of the reservoir grid where there is the intersection of a well with the geological horizon considered, the two-dimensional pattern contains the type of facies estimated geological value for the cell based on logs recorded in the well or, if applicable, based on samples taken from the well.

For example, FIG. 10 (a) shows a seismic horizon 20 of the reservoir grid crossed by four wells 21 -24. Four types of facies were respectively determined on the seismic horizon from the measurements and samplings made in the four wells 21-24. Then, as shown in FIG. 10 (b), the seismic horizon 20 is "flattened" to rearrange the indexes of these cells according to the regular grid I, J by compensating for the deformations of the seismic horizon. This makes it possible to extract the two-dimensional pattern in the form shown in FIG. 10 (c), where the four types of facies raised at the wells are designated by the numbers 1 -4. This two-dimensional pattern, composed of several types of localized facies on the wells, is marked by the position of the regular grid cell which contains the center of gravity of the four wells, marked by an "X" in Figure 10 (c). For each two-dimensional pattern of facies that has thus been extracted from the well data, a scan is performed to count, in the training image, the occurrences of this two-dimensional pattern. For each layer (K index) of the training image, the sweep moves the two-dimensional pattern onto the regular grid. At each of the successive stages of this displacement (for example the step shown in FIG. 10 (d) for one of the layers of the training image), the cell containing the well barycenter X is aligned with a cell reference to the grid of the training image, while the cells composing the two-dimensional pattern are aligned on target cells of the grid of the training image. If the facies types of the training image in these four target cells coincide with those of the two-dimensional pattern, a counter is incremented.

At the end of this operation, the value of the counter is compared with the number of scanned positions, and the ratio between this counter and this number is used to evaluate the ability of the training image to report the correlations between wells. along the seismic horizons as part of the interpuit verification.

[0041] FIG. 11 (a) shows another, more complex example of horizon seismic 30 crossed by six wells 31-36 drilled in the studied area of the basement. FIG. 11 (b) shows the same horizon 30 whose cells have been re-arranged in regular arrangement, which makes it possible to define a two-dimensional pattern composed of six types of facies 41a, 42b, 43b, 44c, 45a, 46a respectively placed on the positions of the six wells 31-36 along the horizon, which positions generally have a center of gravity B. In the notation used for illustration in Figure 1 1 (b), the suffix "a" designates the type of "alluvial plains" facies found on the horizon at wells 31, 35 and 36, the suffix "b" refers to the type of facies "sandy deposit" found on the horizon at wells 32 and 33, and the suffix "c" refers to the type of "channel" facies found at well 44 in the particular example of horizon 30 of Figure 11 (a). This two-dimensional pattern of six types of facies finds an occurrence in the training image layer which is drawn in Figure 11 (c) when centroid B is positioned as shown. Figure 12 shows a possible flow diagram of an inter-well verification procedure in one embodiment of the invention. In a first step 50, the well data, which may include logs and / or rock samples, are analyzed to associate respective facies types with the cells of the reservoir grid located along the wells that have been drilled in the well. area of interest. The analysis can consist of assigning to a particular cell the type of majority facies in the samples taken in this cell, or taking into account physical measurements made in the well. The types of facies thus positioned on the wells can subsequently be used, in step 18 of FIG. 3, to initiate the MPS simulation. Then (step 51), the neighboring wells are grouped together to form one or more groups each composed of M wells (M> 2). The number M can vary from one group to another. The step 51 can be performed manually by a user, or it can be partly automated using a horizontal distance criterion between wells. The count of the occurrences of the two-dimensional patterns is then initialized in step 52, for example by assigning zero values to the NT and N counters and the value 1 to the pattern index q. A pattern M _q is constructed at step 53 of the manner illustrated above with reference to Figures 10 (c) and 11 (b). The pattern M _q is composed of a number M of facies types arranged on a regular 2D grid. These types of facies will then be identified by their relative positions with respect to the centroid B _q , also calculated in step 53, between the positions of the wells of the group considered on the regular 2D grid.

Three nested loops are then executed to scan the drive image for the purpose of detecting the occurrences of the pattern M _q that has just been constructed. These three loops are successively initialized in steps 54, 55 and 56 where the indexes k, j and i are respectively initialized to the values 1, j _m j _n and i _m j _n . The index k numbers the layers of the drive image, while the indexes i, j number cells along both dimensions along the layers. The values i _{m in} and j _min are chosen as a function of the size of the pattern M _q so that it can be superimposed entirely on the considered layer of the training image. At each position of a reference cell i, j, k during scanning, the counter NT is incremented by one unit in step 57. Then (step 58), the position B _q of the center of gravity is aligned with that i, j, k of the current reference cell in the training image, which makes it possible to verify in step 59 whether the pattern M _q matches the types of facies encountered in the position-driving image. corresponding target cells. When there is a match, an occurrence of the pattern M _q is detected and the counter N is incremented by one unit in step 60. After step 60, or in case of mismatch of the pattern in step 59, an end of loop test 61 is performed to compare the index i to its maximum value i _max (depending again on the size of the pattern M _q ). If i <i _max , the index i is incremented by one unit in step 62 and an additional iteration in the inner loop is performed from step 57. When i = i _max in test 61, a second end of loop test 63 is performed to compare the index j to its maximum value j _max (depending on the size of the pattern

Sm). If j <j _max , the index j is incremented by one unit in step 64 and an additional iteration is performed from step 56. When j = j _max in test 63, a third end test of loop 65 is performed to compare the layer index k to its maximum value k _max (number of layers in the training image). Yes ^k < ^k max _> the layer index k is incremented by one in step 66 and an additional iteration is performed from step 55 to scan the new layer. When k = k _max in the test 65, it is examined in the test 67 if all the motives Mq have been searched for in the training image, that is to say if the motive index q is equal to the number of possible patterns Q (equal to the number of groups retained in step 51 multiplied by the number of layers in the reservoir grid). If q <Q, the index q is incremented by one in step 68 and the algorithm returns to step 53 to construct the next pattern. The sweeps of the training image are terminated when q = Q at test 67. At this stage, the numbers N and NT are compared, for example by calculating their ratio a = N / NT at step 69. .

The value of this report provides information on the probability that there are encountered in the drive image 2D facies patterns that can be determined between neighboring wells according to the measurements. The higher values of a validate the training picture from this point of view, while the lower values would encourage the rejection of the training picture. The evaluation may comprise the comparison of a with a predefined threshold, for example of the order of a few%. It may also consist in presenting to the user a score, proportional to a, that this one can take into account together with other scores, resulting for example from intra-well checks of the type of those illustrated with reference to FIGS. and 9.

Alternatively, the ratio a may be calculated for each well group formed in step 51. The comparison threshold for inter-well verification may depend on the number of wells in the group considered. Returning to FIG. 3, when the training image has been validated after the intra- and inter-well verifications of step 15, the next step 17 is to make the connectivity between the wells be honored. During the exploration of a basement region where wells were drilled, or during the production of hydrocarbons in this region, the engineers retrieve dynamic information that provides information on the communications between wells, allowing flows of fluid between them. This information can result from interference testing or simply observing pressures and flows over time. The purpose of step 17 is to take these observations into account in order to initialize the MPS simulation carried out at step 18.

Dynamic observations provide strong information on communications between wells. These communications reveal the existence of flow paths between wells, composed of relatively permeable rocks. A good consideration of the connectivity between wells greatly improves the flow simulations.

FIG. 13 shows, by way of example, two groups of cells comprised in the same layer of the reservoir grid, namely a first group of three cells 80 included in a rectangular assembly 81 and a second group of two cells. 83 in FIG. 13, the reservoir grid layer is shown flat for a constant Kr layer index and variable horizontal indexes Ir, Jr. The three cells 80 of the group included in the rectangular assembly 81 are situated at the intersection of the layer of the reservoir grid with the wells of a group of three communicating wells. Similarly, the two cells 83 of the group included in the rectangular assembly 82 are located at the intersection of the layer of the reservoir grid with the wells of another group of two communicating wells. The identification of such groups of communicating wells is carried out at step 90 shown in FIG. 16. The connectivity is observed between different cells belonging to several wells of the group from the dynamic observations made inside these cells. well when tested or operated. To facilitate reading of Figures 13-15, the cells of a group are shown as belonging to the same layer of the reservoir grid. However, it will be understood that they can be located in different layers, in which case the rectangular sets 81, 82 are generalized to become parallelepiped sets.

[0053] To a g group of cells of the grid reservoir located along several wells and between which hydrodynamic communications were observed, there i _g, j _g the horizontal dimensions of the smallest set parallelepiped 81, 82 containing these cells , and k _g the vertical dimension of this set. In the example of FIG. 13, _g = 4, j _g = 1 1, k _g = 1 for the set 81 and i _g = 3, j _g = 6, k _g = 1 for the set 82 .

For each group of cells of the reservoir grid identified in step 90 of Figure 16 as communicating wells between them, the treatment refers to a rectangular window 84, or parallelepipedic if k _g > 1, which runs through the training image during the scan 91 -104. The rectangular window 84 shown in FIG. 14 corresponds to the rectangular assembly 81 of FIG. 13. In this window 84 are marked the positions 85 which correspond, in the reservoir grid, to the positions of the cells 80 of the group identified in step 90.

The scanning 91 -104 consists of looking for rectangular or parallelepipedic vignettes compatible with the group of cells 80. These compatible thumbnails are related sets of cells of the regular grid that fulfill two conditions:

presenting in the training image, at the marked positions 85 which correspond to the cells 80 of the group, the same types of respective facies as those which have been determined in the wells for these cells 80; and

- include a related subset of cells to which belong those at the marked positions 85 and having in the drive image permeable facies types, i.e., allowing fluid flow between the cells of the group.

[0056] Thus, there is at least two families in facies modeling: a family of relatively permeable facies (1 ^st family, eg channel facies or sandy deposit)), leaving flow of fluids such as oil or natural gas, and a family of less permeable facies (2 ^nd family). These are the facies of the 1 ^st family must be found in the related sub-assemblies to accept compatible thumbnail.

[0057] In Figure 14, the darker squares represent cells having the facies of the ^2nd family, while the other squares represent cells having two types of facies of the 1 ^st family. Reference 86 shows a position where the rectangular window 84 is aligned with a set of cells compatible with the group of cells 80 of Figure 13, while the reference 87 shows a set of cells compatible with the group of cells 83 of Figure 13.

The scan 91 -104 comprises three interleaved loops successively initialized in steps 91, 92 and 93 where the indexes k, j and i are initialized to 1. The index k numbers the layers of the drive image, while the indexes i, j number cells along both dimensions along the layers. At each position i, j, k during scanning, it is examined in step 94 if the related set [i, i + i _g [x [j, j + i _g [x [k, k + k _g [covered in the drive image by the window, here in the form of a rectangular parallelepiped, communicates with each other the cells marked as corresponding to the cells of the group identified in step 90, that is to say if Rectangular parallelepiped has a related subset of cells having permeable facies types in the training image and including the labeled cells. If so, a rectangular parallelepiped compatible thumbnail (or patch) is detected in the training image and the number n of cells of this thumbnail [i, i + i _g [x [j, j + i _g [x [k, k + k _g [whose type of facies is permeable in the training image is counted in step 95.

This number of permeable cells in the compatible thumbnails detected in the training image is the subject of a histogram calculation. The values H (1), H (2), ... of the histogram H are initialized to 0 before step 91. After computing the number n in step 95, the index c receives the value H (n) +1 at step 96 and the current position (i, j, k) of the scan is assigned to an index p ( n, c) locating the position of the ith compatible thumbnail detected in the training image (step 97). The histogram is then updated in step 98 by taking H (n) = c.

After step 98, or when the compatibility test 94 is negative, an end-of-loop test 99 is performed to compare the index i with its maximum value i _m (im + ig is the dimension of the training image along the direction I). If i <i _m , the index i is incremented by one unit in step 100 and an additional iteration in the inner loop is performed from step 94. When i = i _m in test 99, a second end-of-loop test 101 is performed to compare the index j to its maximum value j _m (j _m + j _g is the dimension of the training image along the direction J). If j <j _m , the index j is incremented by one unit at step 102 and an additional iteration is performed from step 93. When j = j _m at test 101, a third end test of loop 103 is performed to compare the layer index k to its maximum value k _m (k _m + k _g is the size of the drive image along the direction K). If k <k _m , the layer index k is incremented by one in step 104 and an additional iteration is performed from step 91. When k = k _m at test 103, the scanning of the training image is completed. The histogram H (n) for the group of wells considered is then completed. It gives the number of compatible thumbnails found during the scan for each number n of permeable cells. It can be expressed equivalently in terms of the volume of permeable facies (by multiplying n by the elementary volume of a cell) or in terms of the proportion of permeable facies (by dividing n by i _g xj _g xk _g ). This histogram can be exploited to select a compatible thumbnail that will be applied to the initialization of the MPS simulation to honor inter-sink connectivity.

In the example of FIG. 16, the scanning of the driving image for a group of connected wells is followed by a step 105 of choosing a number n (or, in an equivalent manner, a volume of permeable facies, or a proportion of permeable facies). This choice is made according to the statistics revealed by the histogram H (n). For example, the average value or the median of the distribution given by the histogram H (n) can be chosen for n. The choice may also take into account a scenario chosen to execute the MPS simulation of step 18. For example, an optimistic scenario may favor the choice of a relatively high number n, whereas a pessimistic scenario may favor the choice. of a relatively low number n.

In step 106, a compatible vignette is chosen, for example by random drawing among the thumbnails detected with the proportion of permeable facies retained in step 105. This may consist of drawing a number c and take the selected thumbnail at the position p (n, c) in the image drive.

To initialize the MPS simulation of step 18 (FIG. 3), an initialization image is constructed on the reservoir grid by placing therein:

- in cells located along a well, the types of facies that were determined by the measurements or samples made in that well;

in the cells situated in the rectangles or parallelepipeds corresponding to a group of connected wells identified in step 90, the types of facies encountered in the training image in the corresponding cells of the vignette selected for this group. FIG. 15 shows part of the initialization image corresponding to FIGS. 13 and 14. This initialization image defined on the reservoir grid comprises (i) a sticker 88 corresponding to the sticker 86 of FIG. on the position of the group of three cells 80 included in the rectangular assembly 81 of FIG. 13 and (ii) another sticker 89 corresponding to the sticker 87 of FIG. 14 keyed on the position of the group of two cells 83 included in FIG. rectangular assembly 82 of Figure 13.

In step 18, the MPS simulation is performed to build, on the reservoir grid, the image modeling the studied area of the basement. The MPS simulation proceeds, in a manner known per se, by propagation from the initialization image by using the training image as a statistical reference.

The MPS simulation and the preliminary preparation of the training images can be implemented using one or more computers. Each computer may comprise a processor type calculation unit, a memory for storing data, a permanent storage system such as one or more hard disks, communication ports for managing communications with external devices, in particular for retrieving data from external devices. satellite images and well data, and user interfaces such as a screen, a keyboard, a mouse, etc.

Typically, the calculations and steps of the above described are executed by the processor or processors using software modules that can be stored as program instructions or code readable by the computer and executable by the processor, on a computer-readable recording medium such as read-only memory (ROM), random access memory (RAM), CD-ROMs, magnetic tapes, floppy disks and optical data storage devices. The embodiments described above are illustrations of the present invention. Various modifications can be made without departing from the scope of the invention which emerges from the appended claims.

Claims

1. A method of preparing a training image for modeling a subsurface area by a multipoint geostatistical technique, the method comprising:

constructing a training image on a first three-dimensional cell grid, the training image giving, for each cell of the first grid, a respective type of geological facies assigned to said cell; and

- verify the representativity of the training image with respect to well data obtained from observations made in several wells that have been drilled in the basement area, the well data referring to a second grid of three-dimensional cells whose morphology is adapted to geological horizons estimated in the subsoil zone, the well data including types of geological facies respectively estimated from observations for cells of the second grid located along the wells ,

characterized in that the representativeness check of the training image with respect to the well data includes an interwell check comprising:

identifying two-dimensional patterns in the well data, each two-dimensional pattern being defined along a respective geological horizon and including, in a cell containing the intersection of a well with said geological horizon in the second grid, the type of estimated geological facies for said cell; and

counting, in the training image, the occurrences of the two-dimensional patterns identified.

The method of claim 1, wherein the interwell check further comprises:

- Validate the training image when the number of counted occurrences of the identified two-dimensional patterns exceeds a threshold.

The method of claim 2, wherein said threshold is expressed in proportion to the number of possible positions of the two-dimensional pattern in the training image.

4. A method according to any one of the preceding claims, wherein each two-dimensional pattern, defined for wells intersecting a geological horizon, is identified by the position of the cell of the second three-dimensional grid containing the barycenter of said wells along said geological horizon. .

The method of claim 4, wherein counting, in the training image, occurrences of a two-dimensional pattern spotted by the position of a centroid cell comprises:

scanning the training image, each step of the scanning placing the two-dimensional pattern on the first grid by aligning the cell containing the barycentre of the wells on a reference cell of the first grid and the cells of the two-dimensional pattern respectively containing the intersections between the wells and the geological horizon on target cells of the first grid; and

The method of any one of the preceding claims, wherein the representativeness check of the training image with respect to the well data further comprises an intra-well check comprising:

- estimating a multipoint statistical measure along each well among the well data;

- estimate the same multipoint statistical measurement along vertical columns of the first grid of cells; and

- compare the two estimated multipoint statistical measures.

The method of claim 6, wherein the estimated multipoint statistical metric is a multipoint density (MPDF) function or a distribution of runs ("distribution of runs").

8. Data processing system adapted to the preparation of a training image for modeling a basement area by a multipoint geostatistical technique, the system comprising computer resources configured to implement a method according to the present invention. any preceding claim.

A computer program for a data processing system, the program comprising instructions for implementing a method according to any one of claims 1 to 7 when executed on a computing unit of the data processing system. data.

A computer readable recording medium, on which is recorded a program according to claim 9.